EP3357195B1

EP3357195B1 - Analysis of network performance

Info

Publication number: EP3357195B1
Application number: EP16775158.5A
Authority: EP
Inventors: Trevor Burbridge; Philip Eardley
Original assignee: British Telecommunications PLC
Current assignee: British Telecommunications PLC
Priority date: 2015-09-30
Filing date: 2016-09-26
Publication date: 2019-11-06
Anticipated expiration: 2036-09-26
Also published as: EP3357195A1; US20180262414A1; CN107852346B; US10320648B2; WO2017055227A1; CN107852346A

Description

Technical Field

The present invention relates to data networks, and to methods and apparatus for analysing performance in respect of data networks. In particular, embodiments thereof relate to ways of analysing or testing network performance in respect of digital data networks such as the Internet, a corporate network, a data centre or a local network using data items such as data packets or other such messages.

Background to the Invention and Prior Art

Analysing networks in order to obtain measurements indicative of network performance can be done using various techniques, including techniques involving active testing (in which traffic is sent over a network specifically for the purpose of conducting tests) and techniques involving passive testing (in which traffic already flowing across a network due to user activity is analysed).
Techniques involving passive testing can show performance of real applications as used by real users, but are generally limited to testing applications and networks being used at a particular time, and can make it hard to compare network performance since the traffic over which tests are being applied varies. Active testing using reference traffic sent across the network does not generally have this disadvantage.
Techniques involving active testing also have problems in what can be tested, however. Typically, active testing techniques either test services themselves (e.g. web page or video performance), or the underlying network. Testing is generally performed from a test-point to a service or to a test-server located in the network. By using multiple test-servers, network operators can get a view of performance across different paths or sub-paths of the network, but it is expensive to deploy and maintain test-servers on a large scale, and this may not give views of networks not under the operator's control unless test-servers are sited within them. There is therefore an interest in using basic network routing equipment to conduct tests, using basic tools such as "traceroute" and "ping".
"Traceroute" is a technique which exploits the feature of Internet Protocol (IP) networks to generate a reply message to the sender of a message when a Time-To-Live (TTL) or hop-limit count expires.
"Ping" is a technique which can be used to test the reachability of nodes in a network, and to measure the round-trip time (RTT) for messages sent from an originating node (such as a computer, server, router, etc.) to a destination node and back. Messages in accordance with the Internet Control Message Protocol (ICMP), referred to as "ICMP probes", "probe messages", or simply "probes", may be sent from a sender acting as a testing-point, generally via one or more intermediate nodes, to a remote network node which, if it is the intended destination of or target for the probe (generally indicated in header information included in the probe), sends an associated probe response message back to the sender, allowing the sender to confirm that the target has been reached and allowing the sender to measure the RTT (also known as latency).
In the present context and below, it will be noted that the word "probe" is generally used in the sense of an "investigation" or one or more "investigative messages", rather than a "sensor". The probes concerned may therefore be one or more packets, or one or more of another type of message sent via a network.
Techniques such as the above are commonly used to determine the nodes located along a network path and also to analyse latency or latency variation between pairs of nodes. Overall latency can determine how far away a node is, while the variation in latency, which may be caused by the filling of network queues, can be used as an indication of network congestion. Such techniques can provide a very fine-grained view of network performance at each node of every network path, allowing performance to be viewed by a network operator even in respect of nodes and paths across networks not under the operator's ownership or control.
A problem with such techniques is that results are not always reliable indicators of network performance. While actual network traffic passing through a node is generally handled in an optimised forwarding element of the node ("fast-path" processing), a "traceroute" response or "ping" will generally be handled by the node's general Central Processing Unit (CPU), and generally involves the generation of a new packet or other such message ("slow-path" processing). Traceroute and ping measurements thus often indicate delays and losses that are not actually experienced by forwarded user traffic.
As a result, previous attempts to determine network performance using basic router functions such as traceroute and ping have often been flawed due to the possibly slow or variable handling of these probes (i.e. probe packets or other messages) by standard network equipment such as routers and other nodes, leading to mis-diagnosis of network problems. Many systems have therefore used specialised testing infrastructure (e.g. dedicated test-servers), but as indicated above, these can generally only give overall end-to-end path performance between the test point and wherever these test-servers are located.
There is thus a need for improved ways of testing network performance which are applicable even when using basic probe techniques such as "traceroute" and "ping" in IP networks.
The "Center for Applied Internet Data Analysis" ("CAIDA") has developed a tool called "Scamper" for use in a project referred to as the "Archipelago" project. This is intended to allows bulk traceroute and ping measurements. They have published the following papers:

"Challenges in Inferring Internet Interdomain Congestion" by M. Luckie, A. Dhamdhere, D. Clark, B. Huffaker, & K. Claffy, Internet Measurement Conference (IMC), November 2014, pages 15-22, which is available online at: https://www.caida.ora/publications/papers/2014/challenaes inferring interdomain conaestion/
and "Measurement and Analysis of Internet Interconnection and Congestion" by D. Clark, S. Bauer, K. Claffy, A. Dhamdhere, B. Huffaker, W. Lehr, & M. Luckie, Telecommunications Policy Research Conference (TPRC), September 2014, which is available online at: https://www.caida.ora/publications/papers/2014/measurement analysis internet interconnection/

These papers consider how data can be used to infer congestion, particularly between network domains, and discuss how to analyse the data to detect network problems.
Referring to other prior art citations, US2010103822 ("Montwill ") relates to network assessment and fault isolation, and in particular to identifying problem nodes between two end-points using traceroute packets.
US2007274227 ("Rauscher ") relates to the measurement of network latency in a packet switched network, and in particular to methods for traversing a network to calculate network latency.
EP2557731 ("Huawei ") relates to methods and systems for an intermediate node to locate a fault independently, where the intermediate node serves to forward a packet between a first end node and a second end node.
EP1206085 ("Infonet ") relates to methods and apparatus for automated service level agreements.
An IETF Network Working Group Internet Draft entitled "A Round-trip Delay Metric for IPPM" dated November 1998 and authored by G. Almes, S. Kalidindi and M. Zekauskas defines a metric for round-trip delay of packets across Internet paths.
A "Tech Notes" publication from Cisco Systems entitled "Understanding the Ping and Traceroute Commands" (http://www.cisco.com/image/gif/paws/12778/ping_traceroute.pdf) dated January 2010 illustrates the use of the ping and traceroute commands and, with the aid of some debug commands, captures a more detailed view of how these commands work.
US2010/315958 ("Luo et al ") relates to methods and apparatus for measuring network path quality in a non-cooperative manner, and involves sending a probe consisting of probe data packets to a remote node and receiving a response consisting of at least one response data packet therefrom.

Summary of the Invention

The CAIDA papers discussed above discuss how to analyse data to detect network problems, but do not consider problems associated with specific nodes themselves having poor or variable response to network probes, let alone methods to deal with such problems.
Embodiments of the invention are based on the realisation that measurements made from sending a probe message such as a "Ping" message from a "testing" network node to a "target" network node, which are supposed to be indicative (at least primarily) of the performance of the network in respect of the path between the respective nodes, may also be influenced unduly by the time taken by the target node itself to perform local, on-board, "slow-path" processing of the probe message. Such "slow-path" processing generally only occurs in respect of probe messages where a node is the "target" node - if the same node is merely an intermediate node forwarding a probe message to a subsequent node which is the target node, it will generally only perform "fast-path" processing in respect of that probe message (i.e. to inspect its header and forward it). The present inventors have realised that this provides an opportunity to isolate and estimate the effect of a particular node's "slow-path" processing, and if it is estimated to be having a particularly damaging effect on the reliability of measurements where it is the target node, network performance analysis can be based (at least primarily) on other nodes whose own effect on probe measurements is less damaging.
According to a first aspect of the invention, there is provided a method of analysing network performance of a network comprising a plurality of network nodes, the method comprising:

obtaining, in respect of a first target node, a plurality of probe measurements, the probe measurements in respect of the first target node including:
- at least one probe measurement resulting from one or more probe test-messages being sent via a network path from a testing node to said first target node and one or more associated probe response-messages triggered by receipt and local processing at said first target node of the one or more probe test-messages being received via a network path from the first target node by the testing node; and
- at least one probe measurement resulting from one or more probe test-messages being sent via a network path from a testing node via the first target node to a subsequent target node and one or more associated probe response-messages triggered by receipt and local processing at said subsequent target node of the one or more probe test-messages being received via a network path from the subsequent target node by the testing node;
the respective probe measurements each relating to one or more network performance characteristics in respect of the network paths taken by the respective probe test-messages and the probe response-messages associated therewith and being dependent also on the local processing of the respective probe test-messages at the respective target nodes;
comparing at least one of said one or more probe measurements resulting from one or more probe test-messages being sent to said first target node with at least one of said one or more probe measurements resulting from one or more probe test-messages being sent via the first target node to a subsequent target node, and in dependence on the comparison, assigning a weighting in respect of at least one of said target nodes;
determining a network performance analysis measure according to a predetermined function dependent on at least one probe measurement in respect of at least one of said target nodes and on the weighting assigned in respect of said at least one target node.

Preferred embodiments may be used to identify whether a particular target node is likely to be providing unreliable results by comparing results obtained by sending probes to it with results obtained by sending similar probes via the first target node to a subsequent (further-away) target node. If the results in respect of the further-away node indicate a lower round-trip time than those in respect of the nearer node (despite this being a shorter distance along the path to the further-away node), for example, this is a strong indication that the results in respect of the nearer node are being unduly influenced by internal processing within the nearer node itself, and are thus unreliable, so should be discounted or given lower weight when analysing network performance.
The predetermined function may be chosen such that the overall performance analysis measure may be based on or most strongly based on the probe measurement(s) obtained in respect of a node, or a number of nodes, from which the "best" measurement(s) was (were) obtained, or on one or more probe measurements subsequently obtained in respect of that target node or those target nodes. It will be understood that the meaning of the term "best" will depend on the type of measurement being made, but in general, in relation to some types of measurements such as "time" or "variability" measurements, the lowest will generally be deemed the best, whereas in relation to others, e.g. reliability or data volume measurements, the highest will generally be deemed the best.
The "return" network path taken by a probe response-message of a particular probe attempt will usually be the reverse of the "outward" network path taken by the associated probe test-message, but this will not necessarily be the case, and may not even be a factor under the control of entities in control of the testing node or the target node in question.
According to preferred embodiments, the step of obtaining probe measurements in respect of the first target node may comprise obtaining respective probe measurements resulting from probe test-messages being sent via the first target node to a plurality of subsequent target nodes and associated probe response-messages triggered by receipt and local processing at the subsequent nodes of the probe test-messages being received from the subsequent target nodes.
According to preferred embodiments, respective weightings may be assigned in respect of a plurality of the target nodes such that the network performance analysis measure is dependent on one or more probe measurements in respect of one or more of the respective target nodes, e.g. the or each target node to which a positive weighting has been assigned. Correspondingly, respective weightings may be assigned in respect of a plurality of the target nodes such that the network performance analysis measure is independent of one or more probe measurements in respect of one or more of the respective target nodes, e.g. any target node to which a zero weighting has been assigned.
The weightings may be "1" (i.e. full, or 100%) and "0" (i.e. zero, or 0%), in order either to include or exclude measurements in respect of particular target nodes in the overall analysis measure, or may be on a scale from 1 to 0 in order to allow the overall analysis measure to be influenced to a greater/lesser extent by measurements made in respect of target nodes deemed to be providing more/less reliable individual measurements.
According to preferred embodiments, the predetermined function may be such that the network performance analysis measure is dependent on one or more probe measurements obtained in respect of each of a plurality of the respective target nodes, e.g. the or each target node to which a positive weighting has been assigned. Correspondingly, the predetermined function may be such that the network performance analysis measure is independent of one or more probe measurements obtained in respect of each of a plurality of the respective target nodes, e.g. any target node to which a zero weighting has been assigned.
According to preferred embodiments, probe measurements may be obtained and compared in respect of a plurality of the target nodes whereby to determine weightings in respect thereof.
According to preferred embodiments, subsequent probe measurements may be obtained in respect of a subset of one or more target nodes selected from the plurality of target nodes, the subset comprising one or more target nodes selected in dependence on the comparison.
According to preferred embodiments, one or more of the probe measurements may result from a plurality of probe test-messages being sent via a network path from a testing node to the same target node at different times.
According to preferred embodiments, the step of obtaining a plurality of probe measurements in respect of the first target node may comprise obtaining a plurality of probe measurements resulting from probe test-messages being sent to the same target node, and the step of comparing may involve comparing the plurality of probe measurements resulting from probe test-messages being sent to the same target node with at least one of the one or more probe measurements resulting from one or more probe test-messages being sent to the other target node. The plurality of probe measurements resulting from probe test-messages being sent to the same target node may result from probe test-messages being sent to the same target node at different times.
According to preferred embodiments, the probe measurements obtained in respect of the respective target nodes may include measurements in respect of one or more of the following:

one or more characteristics associated with response time (e.g. round-trip response time to a "ping" message);
one or more characteristics associated with communication speed;
one or more characteristics associated with communication delay and/or delay variation;
one or more characteristics associated with communication volume;
one or more characteristics associated with reliability;
one or more characteristics associated with data loss;
one or more characteristics associated with communications quality;
one or more characteristics associated with security;
one or more characteristics associated with service usage;

According to a second aspect of the invention, there is provided apparatus configured to perform a method in accordance with the first aspect.
The apparatus may comprise one or more network nodes such as routers, or one or more processors associated with one or more such nodes, for example.
According to a third aspect of the invention, there is provided a computer program element comprising computer program code to, when loaded into a computer system and executed thereon, cause the computer to perform the steps of a method according to the first aspect.
The various options and preferred embodiments referred to above in relation to the first aspect are also applicable in relation to the second and third aspects.
Preferred embodiments of the invention are able to utilise existing tests based upon existing functionality found in IP routers. Traceroute and ping are available as client tools along with more comprehensive tools such as scamper that allow bulk automation of traceroute and ping tests. Traceroute exploits the feature of IP networks to generate a reply to the sender when a Time-To-Live (TTL) expires. Ping reflects special ICMP probes to a network node back to the sender in order to test the round-trip-time (also known as latency).
Preferred embodiments of the invention are able to use such tools to produce individual test results for routers along a path from the test point. Such test results may record the time at which the test was executed, the IP node being tested, and performance data such as whether a response was returned along with the delay (or round-trip time) taken to respond.
These individual results may then be used to perform an analysis of the performance in respect of each of a number of nodes. The performance considered is commonly the overall (average) latency to reach the node as well as variation in the latency which might indicate congestion. The techniques developed by CAIDA discussed above can be used to look for diurnal variation in the latency which would be expected from peak-time network usage, however such diurnal variations can also exist in a node's response to traceroute and ping even when the network is not congested.

Brief Description of the Drawings

Preferred embodiments of the present invention will be described with reference to the appended drawings, in which:

Figure 1: shows a network node and two neighbouring nodes;
Figure 2: shows several network nodes on a path;
Figure 3: illustrates "fast-path" and "slow-path" processing at different nodes; and
Figure 4: is a flow-chart illustrating a method according to a preferred embodiment.

Description of Preferred Embodiments of the Invention

With reference to the accompanying figures, methods and apparatus according to preferred embodiments will be described.
Before describing preferred embodiments of the invention, the issue of "slow-path" and "fast-path" processing in network nodes such as routers, referred to briefly above, will be explained in more detail with reference to Figures 1, 2 and 3.
Referring to Figure 1 , this shows a Network Node 10 and two Neighbouring Nodes 10' and 10", which are referred to respectively as being an "upstream" Node 10' and a "downstream" Node 10". Network Node 10 is shown expanded and in detail, with individual/internal functional modules shown (noting that these may be implemented as software or hardware modules, and noting that the separation between them may in fact be functional, rather than spatial), in order to assist with an explanation of the main functions thereof. The two neighbouring nodes 10' and 10" would generally have similar or corresponding individual/internal functional modules, but these are not shown in order to avoid unnecessarily complicating the figure.
It will be understood that the terms "upstream" and "downstream" are comparative, and depend on the role a particular node is playing in relation to a particular exchange of data - a first node can be upstream of a second node in relation to the path taken by one packet and downstream of the second node in relation to the path taken by another packet - but for the present discussion, these terms are used in relation to scenarios where:

(i) upstream node 10' is sending data (e.g. one or more probe messages, comprising one or more data packets) along a path via node 10 to "downstream" network node 10", in which case the function of node 10 is to receive data from upstream node 10' and forward it on to downstream node 10"; and
(ii) upstream node 10' is sending data to node 10 itself and receiving a response therefrom.

In Figure 1 , the solid arrows (H1, h1, h2, H2) relate to scenario (i), and indicate the one-way path taken by data sent from upstream node 10' having downstream node 10" as its (intended) destination address (as indicated in its header or otherwise), that reaches and is forwarded by node 10 on its path from upstream node 10' to downstream node 10". Of these arrows, those with a capital "H" (i.e. H1 and H2) indicate external hops on the path, i.e. the hop H1 between nodes 10' and 10, and the hop H2 between nodes 10 and 10". Those with a lower-case "h" (i.e. h1 and h2) indicate internal processing paths within node 10 for data received from node 10' by a first input/output (I/O) interface 101 of node 10 and passed to the forwarding processor 103 of node 10 (h1), then passed from the forwarding processor 103 of node 10 (h2) to a second I/O interface 109 of node 10 which presents the data for forwarding from node 10 on to node 10".
As indicated above, data packets in an IP network generally indicate their destination IP address (as well as their source address and other information) in their IP header, which would generally be used in relation to a "ping" test. For a traceroute request, the header may (also) indicate the Time-To-Live (TTL) hop count, but in this scenario, the target IP address might not be the same as the traceroute destination/target.
The dotted arrows (h3, h4, h5 and H3, and ignoring h1/3 and h4/5 for the time being) relate to scenario (ii), and together with solid arrows H1 and H3 indicate the two-way path taken when data is sent from upstream node 10' having node 10 as its (intended) destination address, when that data reaches node 10 and is processed there, and when an associated response is sent by node 10 back to upstream node 10'. Again, arrows with a capital "H" (i.e. H1 and H3) indicate external hops on the path, i.e. the hop H1 between nodes 10' and 10 (also performed in scenario (i)), and the hop H3 back from node 10 to node 10'. Those arrows with a lower-case "h" (i.e. h1, h3, h4 and h5) indicate internal processing paths within node 10 for data received from node 10' by I/O interface 101 of node 10 and passed to the forwarding processor 103 of node 10 (h1, common to scenario (i)), which is found by node 10 to be targeted at node 10 itself, and which is therefore passed for processing by the CPU 105 of node 10 (h3), processed there, passed back to the forwarding processor 103 of node 10 (h4), passed from there back to I/O interface 101 of node 10 (h5) and presented for forwarding from node 10 back to node 10' (H3). The CPU 105 is shown as having an associated memory 107, in which it may store information such as routing tables, the IP addresses of the interfaces, etc.
Thus, in relation to scenario (i), if node 10 is serving as an intermediate node between an upstream node 10' and a downstream node 10", and is therefore required merely to forward data such as a probe message from node 10' on to node 10", the path taken by the data is: $H 1 - > h 1 - > h 2 - > H 2$
In relation to scenario (ii), however, if node 10 is serving as the target or destination node for data such as a probe message sent to it from upstream node 10', and is therefore requested to send a probe response to the upstream node 10' (so is therefore required to process the probe message in its CPU 105 before sending a response message back to node 10', the path taken by the data is: $H 1 - > h 1 - > h 3 - > h 4 - > h 5 - > H 3$
(NB In relation to the presently-described embodiment as described above, the I/O interfaces 101 and 109 of Node 10 serve simply as interfaces whose function is to forward data received from an external node (e.g. Node 10' or Node 10") to the forwarding processor 103, and to forward data received from the forwarding processor 103 to an external node (e.g. Node 10" or 10'). It is the forwarding processor 103 whose function it is to inspect received data and, in the case of most data packets, including messages such as "ping" requests, determine (e.g. from reading a "destination address" indication in a header of a packet or other such message) whether the node itself is the intended destination for the data (in which case the data is passed to the CPU 105) or whether another node is the intended destination for the data (in which case the data is passed to the appropriate I/O interface for forwarding). (For "traceroute" tests, the issue may be whether a packet's header indicates an expired TTL, rather than whether the packet has reached its intended destination, however.) In any case, it will be noted however that curved dotted arrows referred to as "h1/3" and "h4/5" are shown in Figure 1. These illustrate the situation in relation to alternative embodiments, in which the function of inspecting received data and determining whether the node itself is the intended destination for the data or whether another node is the intended destination for the data may be performed by the I/O interface receiving the data. In such embodiments, data requiring processing by the CPU 105 may be passed to it internally from the I/O interface 101 along the internal processing path indicated by curved dotted arrow h1/3, without passing through forwarding processor 103, and response data may be passed internally from the CPU 105 back along the path indicated by curved dotted arrow h4/5 to the I/O interface 101, again without passing through forwarding processor 103. The issue of "fast-path" and "slow-path" processing discussed above and below is applicable in relation to both types of embodiment.)
An important issue to note in relation to the above is that when forwarding a probe (such as a "traceroute" or "ping" message), a normal node (as opposed to a dedicated test-server) typically handles this in the same fashion as it handles other packets (NB it is said to be processed on the "fast path", where the word 'path' here refers to an 'internal processing path within the node', rather than to a network path); the node is optimised for forwarding packets and the operation may be carried out entirely in hardware. However, when such a node responds to a probe (for instance, because the TTL of a traceroute packet has reached zero, or because the node is the target or intended destination of a "ping" packet), then the node is said to process the packet on the "slow path" (the word 'path' again referring to an 'internal processing path within the node'); the "response" operation involves the node's CPU, and generally includes generation of a new packet/message which is sent back towards the source of the probe.
As a result of this, measurements of a characteristic relating to performance along a network path (e.g. round-trip response time or delay, variation in round-trip response time or delay, etc.) desired to be measured using one or more probe messages can be distorted by the "slow-path" processing of the probe message(s) on the probe's "target node" itself due to its internal processing, and in particular due to the speed, congestion-state, reliability or another characteristic of the CPU of the target node itself. If the performance measurement desired is one indicative of the performance state of the path between the testing node and the target node, distortion to such measurements being unduly influenced by the "slow-path" processing of the target node itself being unduly slow is generally undesirable.
It can thus be desirable to identify nodes causing (or deemed likely to be causing) significant distortion to individual probe measurements by virtue of their own "slow-path" processing and/or to identify nodes causing (or deemed likely to be causing) the least or most distortion, and therefore allow overall network analysis to be based at least primarily on probe measurements from one or more nodes found (or deemed likely) to be causing the least distortion, and possibly to allow measurements likely to have been more distorted to be removed or ignored from the overall network analysis.
In the interest of simplifying the explanation of the above issues, Figure 1 only illustrates situations involving the minimum numbers of nodes necessary, but it will be understood that paths across networks may involve several nodes and hops therebetween, and networks are generally branched, so other routes between nodes may also be possible.
Further, while Node 10 is shown as having two I/O interfaces, a first one (101) for receiving data from and sending data to Node 10', and second one (109) for receiving data from and sending data to Node 10", a node may just have one interface for all sending and receiving, one interface for sending and another for receiving, several interfaces, one for each of a number of neighbouring nodes, or another such arrangement of interfaces.
With reference to Figure 2 , for example, which shows several network nodes on a path, it will be understood that a path across a network may generally have several intermediate nodes between an original sender (e.g. Test Source S, 20) of a message such as a probe message and the target for that data. For example if Node E, 20e, is the target node for a probe message 22e from Test Source S, and data travelling from Test Source S to Node E, is received and forwarded respectively by Nodes A, B, C, and D (20a, 20b, 20c, 20d) before reaching Node E, each of Nodes A, B, C and D acts as an intermediate node in respect of the probe message 22e from Test Source S to Node E, and may also act as an intermediate node in respect of as associated response message 24e from Node E back to Test Source S (although it should be noted that this "return" path may not necessarily be the reverse of the "outward" path). In such a scenario, the messages would be subjected to "fast-path" processing by each of Nodes A, B, C and D, and to "slow-path" processing only by Node E.
It will also be understood that while Test Source S is shown as sending probe messages 22a, 22b, 22c, 22d and 22e to each of Nodes A, B, C, D and E (i.e. each acting as the Target Node in respect of the probe message sent thereto) and receiving associated response messages 24a, 24b, 24c, 24d and 24e therefrom, any of the nodes shown may act as a Test Source and send probe messages to and receive associated response messages from any of the other nodes, at the same time or at different times, which may travel along any path linking the nodes, in either direction. Whether a particular node is involved in a particular probe attempt as a Test Source, as an intermediate node, or as a Target Node will determine whether its role will generally involve sending, forwarding, or receiving and responding to a probe message, which in turn will generally determine whether the particular node will process the probe only via its "fast-path" processing route (in order to send or forward the probe message downstream to another node, or forward a response message back towards the Test Source in question) or (additionally) via its "slow-path" processing route (in order to process the probe message in its own CPU, create a response message, and present this for transmittal back upstream towards the Test Source in question).
Figure 3 illustrates how "fast-path" and "slow-path" processing may be performed at different nodes depending on whether the nodes are serving as the Target Node in respect of a particular probe or merely as an Intermediate Node on a path between a Test Source and a Target Node. For simplicity, it simply illustrates two probe messages being sent from a Test Source S, 30, one of which is sent with Node A, 30a as its Target Node, and the other of which is sent with Node B, 30b as its Target Node, along a path on which Node A, 30a is an Intermediate Node. It will be understood that in different embodiments, the respective probe messages may be sent at the same time by Test Source S, 30, or at different times, that multiple messages may be sent to each Target Node, and that other nodes may also be involved as Test Sources, Intermediate Nodes and Target Nodes, at the same or at different times. This simple example will be used in order to explain how comparisons of different probe measurements made (using "ping", "traceroute", or other such probing techniques) may be made and used.
The individual functional modules of the nodes 30, 30a and 30b (i.e. the I/O Interfaces, the Forwarding Processors, the CPUs and their associated memories) are shown in abbreviated form where applicable, but as these have been explained earlier with reference to Figure 1 , they are not individually numbered in Figure 3 so as not to over-complicate the figure. Dotted arrows have been used in Figure 3 in order to show the paths concerned (i.e. the external hops between the nodes as well as the internal processing paths within the nodes) which are traversed in respect of the respective probes (i.e. the respective "outward" paths taken by the respective probe test-messages from the Test Source S to the respective Target Nodes, and the respective "return" paths taken by the respective probe response-messages from the respective Target Nodes back to the Test Source S), but again, these are not labelled as it is thought that this would make the figure unnecessarily complex.
With preferred embodiments of the invention, performance measurements obtained from different probe messages being sent on different paths and/or between different Test Sources and/or Target Nodes are compared in such a way as to indicate which measurements have been, or are likely to have been, unduly influenced by unduly slow, unreliable, misleading or otherwise poor "slow-path" processing in the node acting as the Target Node for the probe message in question. The results of such comparisons can be indicative of which node or nodes is/are responding in such a way as to provide genuine or realistic indications of performance on the network path in question.
Referring to the left-hand side of Figure 3 , consider that a probe from Test Source S to Node A involves the following elements being traversed in the course of allowing a single probe measurement to be made:

(i) the "outward" network path from the originating test-point, Test Source S, to Node A (which itself comprises the dotted arrows within Test Source S from its CPU and on to its I/O interface to Node A, and the dotted arrow from the I/O interface of Test Source S to the I/O interface of Node A, but this sub-division is not of importance because the parts are common to the left-hand side and the right-hand side of the figure);
(ii) the ingress interface on Node A from Test Source S;
(iii) the response processing in respect of the probe within Node A (which itself is represented by the dotted arrows within Node A from its I/O interface via its Forwarding Processor to its CPU, and back via its Forwarding Processor to its I/O interface);
(iv) the egress interface on Node A back towards Test Source S; and
(v) the "return" network path to the Test Source S.

Simplifying the terminology, we can say the following: $Overall Node A measurement = Return path from S to A + Probe response of A$
where the contribution referred to as "Return path from S to A" to the overall Node A measurement includes the following four sub-contributions:

(i) a sub-contribution relating to the "outward" network path from the Test Source S to Node A;
(ii) a sub-contribution relating to the ingress interface on Node A;
(iv) a sub-contribution relating to the egress interface on Node A back towards Test Source S; and
(v) a sub-contribution relating to the "return" network path from Node A to Test Source S.

It will be noted that the only part of the overall measurement not also forming part of the "Return path from S to A" contribution is that relating to (iii) above, i.e. the response processing in respect of the probe within Node A itself.
Now, with reference to the right-hand side of Figure 3 , consider a probe from the same Test Source S to subsequent Node B, 30b, sent via Node A, 30a. This measures the same elements as the probe to Node A except for the "slow-path" response processing in respect of the probe within Node A itself (as the probe to Node B is merely forwarded across Node A, via its "fast path" processing, rather than processed by the CPU of Node A via its "slow path" processing, but in addition, it now also includes the network path between Node A and Node B ("outward" and "return" contributions, as represented by the dotted arrows between Node A and Node B) and the probe response of Node B itself (represented by the dotted arrows within Node B from its I/O interface via its Forwarding Processor to its CPU, and back via its Forwarding Processor to its I/O interface).
Using the same convention as above, the contributions of overall measurements can be regarded as follows: $Overall Node B measurement = Return path from S to B + Probe response of B$
$Overall Node B measurement = Return path from S to A + Return path between A and B + Probe response of B$
$Overall Node B measurement = Overall Node A measurement - Probe response of Node A return path between Node A and Node B + Probe response of Node B$
and $Overall Node B measurement - Overall Node A measurement = Return path between Node A and Node B + Probe response of Node B - Probe response of Node A$
It can be seen from the above that the performance (which may be measured in terms of speed, throughput, reliability, consistency or any of a variety of other types of characteristic) seen when Node B is the Target Node will not always be slower, longer, smaller, larger, or otherwise worse (depending on the type of performance characteristic in question) than that seen when Node A is the Target Node (e.g. if the Node A probe response is unduly slow). However, any network performance of the path to Node A must be included in the performance seen at Node B. Thus, while it may not be possible to isolate clearly or completely the network performance to Node A, it is possible to identify node measurements that strongly appear to be compromised by the probe response time of Node A if, for example, a round-trip time measurement in respect of a probe from Test Source S to Node A has a greater RTT than a concurrent measurement in respect of a probe from the same Test Source S via Node A to subsequent Node B.

Identification and Treatment of Suspect Node Measurements

With preferred embodiments of the invention, the performance as indicated by one or more measurements made in respect of a first node acting as a "Target Node" is compared to the performance as indicated by one or more measurements made in respect of one or more subsequent nodes (on a path via the first node) as "Target Nodes". The overall motivation is that if the results obtained are indicative of genuine forwarding performance on the network path in question, then any degradation should also be seen at subsequent nodes.
Methods in accordance with a preferred embodiment will now be described with reference to Figure 4 , which is a flow-chart illustrating how comparisons made between probe measurements in respect of a Target Node and one or more subsequent Target Nodes can be used to analyse network performance, and in particular to identify and reduce or remove reliance on questionable node data (Implementation A).
For this explanation, we will consider measurements of round-trip time (RTT) obtained by "ping" probes, but it will be understood that others characteristics may be measured using other probe techniques. We will consider an embodiment in which batches of probe measurements may be collected, which may include measurements relating to several possible "First Target Nodes" and their associated "Subsequent Target Nodes", but it will be understood that in other embodiments, the process may simply perform the analysis in relation to one "First Target Nodes" and its associated "Subsequent Target Node(s)" at a time.
In a preferred embodiment, probe measurements are collected (step s40), one or more being in respect of each of at least two Target Nodes (such as Nodes A and B in Figure 3), one of which (e.g. Node A) is on a network path between the common Test Source (e.g. Node S) and the other (e.g. Node B). If more than one measurement in respect of each node is available, the average, the total, the highest, the lowest, the variation or standard deviation, a quantile (e.g. the 95th percentile) of the results or some other appropriate function of those results available may be used, depending on the characteristic in question.
Of the possible Target Nodes, a first Target Node is selected (step s41), which will be Node A in the simple case shown in Figure 3, and a subsequent Target Node is selected (step s42), which will be Node B in the simple case shown in Figure 3. The probe measurements from the respective Target Nodes (A and B) are then compared (step s43).
If it is found at step s44 that measurements in respect of Target Node A give a lower RTT than those from the subsequent Target Node B, it can be concluded (at step s45) that the Node A measurements are at least possibly valid (i.e. there is no clear suggestion that they are questionable). In this case, a positive weighting, which may be "1" or a value between "0" and "1", may be assigned to Node A or to the measurement(s) obtained therefrom, indicating that Node A measurements can be used in the overall determination of a network performance measure. If it is found (at step s46) that measurements from any other subsequent Target Nodes to Node A are available, the process may return to step s42 and perform a similar comparison between measurements from these (as subsequent Target Nodes) and Node A as the Target Node, in order to gain more evidence as to whether internal processing within Node A is affecting the RTT measurements to it unduly. If it is found (at step s46) that measurements from no other subsequent Target Nodes are available, the process may continue to step s47, at which an overall Network Performance Measure can be determined based on measurements already or subsequently obtained with Node A as the Target Node. The process may then return to step s40 or s41 and be repeated with new measurements or in respect of a different "First Target Node" (i.e. in order to perform a corresponding investigation in respect thereof).
If it is found (at step s44) that the measurements in respect of Target Node A give a higher RTT than those from the subsequent Target Node B, it can be concluded (at step s48) that the Node A measurements are questionable. While this gives no specific suggestion that the Node B measurements are valid, it indicates that they are less questionable than those from Node A. A positive weighting can thus be assigned to Node B or to the measurement(s) obtained therefrom, indicating that Node B measurements can be used in the overall determination of an overall Network Performance Measure. As before the weighting may be "1" or a value between "0" and "1", depending on whether it is desired to determine the overall Network Performance Measure (at step s49) on the basis of measurements from the "best" node (e.g. Node B) alone or on the basis of a weighted sum of measurements from nodes having a positive weighting. As before, the process may then return to step s40 or s41 and be repeated with new measurements or in respect of a different "First Target Node".
A single node measurement can be compared to one or more subsequent node measurements. If the node has little or no probe response delay then all the subsequent nodes should have more delay and/or probe failures, for example. While one or more of the subsequent nodes may also have questionable response times, comparing a particular node's measurement against a greater number of subsequent nodes will generally increase the potential to detect nodes with significant probe response delays and allow them to be ignored or weighted so as not to have a significant effect on overall network performance measures.
This approach can filter out, weight, or apply confidence levels to node measurements that clearly have a degree of node probe delay and are therefore suspect for use in analysing the network path performance.
When considering whether a result is really higher than another result, the error or confidence of the value should preferably also be considered. For example it might be decided to discard a node's measurements if it is determined with a 90% confidence level (for example) that its delay exceeds that of subsequent nodes, even though the average of the measurement samples is actually higher.

Time Analysis

Simply comparing the average delay or probe failures between two nodes and discarding measurements that are too high in early nodes can still leave measurements that are still compromised by the node probe response delay. For example, Node A may have very variable probe response delay. Averaged over time, the delay at one node may look like it is lower than that at subsequent nodes. However, at a single point in time it might be possible to see that the delay on Node A actually far exceeds the delay at the same time on subsequent nodes. In this case, the measurements at Node A can therefore still be concluded to be questionable for network performance analysis.
In order to filter out more suspect node measurements, the measurements at each node can be divided in temporal space. For example the measurements could simply be split into peak-time and off-peak network usage periods. If a node had a higher peak or off-peak measurement than subsequent nodes then it could be considered questionable. With enough results each node can be further sub-divided, for example into hours of the day (e.g. 8-9pm), or simply within buckets along a continuous time axis (e.g. 8-9pm, 20th August 2015). It may be decided that for a node's measurements to be considered valid, it should be found (to a predetermined degree of confidence) that the delay and/or failure level in each division of the temporal space is higher at all subsequent nodes.
Insofar as embodiments described are implementable at least in part using a software-controlled programmable processing device, such as a microprocessor, digital signal processor or other processing device, data processing apparatus or system, it will be appreciated that a computer program for configuring a programmable device, apparatus or system to implement the foregoing described methods is envisaged as an aspect of the present invention. The computer program may be embodied as source code or undergo compilation for implementation on a processing device, apparatus or system or may be embodied as object code, for example.
Suitably, the computer program is stored on a carrier medium in machine or device readable form, for example in solid-state memory, magnetic memory such as disk or tape, optically or magneto-optically readable memory such as compact disk or digital versatile disk etc., and the processing device utilises the program or a part thereof to configure it for operation. The computer program may be supplied from a remote source embodied in a communications medium such as an electronic signal, radio frequency carrier wave or optical carrier wave. Such carrier media are also envisaged as aspects of the present invention.
It will be understood by those skilled in the art that, although the present invention has been described in relation to the above described example embodiments, the invention is not limited thereto and that there are many possible variations and modifications which fall within its scope.
The scope of the present invention includes any novel features or combination of features disclosed herein. The applicant hereby gives notice that new claims may be formulated to such features or combination of features during prosecution of this application or of any such further applications derived therefrom. In particular, with reference to the appended claims, features from dependent claims may be combined with those of the independent claims and features from respective independent claims may be combined in any appropriate manner and not merely in the specific combinations enumerated in the claims.

Claims

A method of analysing network performance of a network comprising a plurality of network nodes (30, 30a, 30b), the method comprising:
obtaining, in respect of a first target node (30a), a plurality of probe measurements, the probe measurements in respect of the first target node including:
- at least one probe measurement resulting from one or more probe test-messages being sent via a network path from a testing node (30) to said first target node (30a) and one or more associated probe response-messages triggered by receipt and local processing at said first target node of the one or more probe test-messages being received via a network path from the first target node by the testing node; and

- at least one probe measurement resulting from one or more probe test-messages being sent via a network path from a testing node (30) via the first target node (30a) to a subsequent target node (30b) and one or more associated probe response-messages triggered by receipt and local processing at said subsequent target node of the one or more probe test-messages being received via a network path from the subsequent target node by the testing node;
the respective probe measurements each relating to one or more network performance characteristics in respect of the network paths taken by the respective probe test-messages and the probe response-messages associated therewith and being dependent also on the local processing of the respective probe test-messages at the respective target nodes; and

comparing at least one of said one or more probe measurements resulting from one or more probe test-messages being sent to said first target node with at least one of said one or more probe measurements resulting from one or more probe test-messages being sent via the first target node to a subsequent target node;

characterised in that the method further comprises assigning a weighting in respect of at least one of said target nodes in dependence on the comparison; and

determining a network performance analysis measure according to a predetermined function dependent on at least one probe measurement in respect of at least one of said target nodes and on the weighting assigned in respect of said at least one target node.
A method according to Claim 1 wherein the step of obtaining probe measurements in respect of the first target node (30a) comprises obtaining respective probe measurements resulting from probe test-messages being sent via the first target node to a plurality of subsequent target nodes and associated probe response-messages triggered by receipt and local processing at said subsequent target nodes of the probe test-messages being received from the subsequent target nodes.
A method according to Claim 1 or 2 wherein respective weightings are assigned in respect of a plurality of said target nodes (30a, 30b) such that said network performance analysis measure is dependent on one or more probe measurements in respect of one or more of said respective target nodes.
A method according to Claim 1, 2 or 3 wherein respective weightings are assigned in respect of a plurality of said target nodes (30a, 30b) such that said network performance analysis measure is independent of one or more probe measurements in respect of one or more of said respective target nodes.
A method according to any of the preceding claims wherein the predetermined function is such that the network performance analysis measure is dependent on one or more probe measurements obtained in respect of each of a plurality of said respective target nodes (30a, 30b).
A method according to any of the preceding claims wherein the predetermined function is such that the network performance analysis measure is independent of one or more probe measurements obtained in respect of each of a plurality of said respective target nodes (30a, 30b).
A method according to any of the preceding claims wherein probe measurements are obtained and compared in respect of a plurality of said target nodes (30a, 30b) whereby to determine weightings in respect thereof.
A method according to any of the preceding claims wherein subsequent probe measurements are obtained in respect of a subset of one or more target nodes selected from the plurality of target nodes (30a, 30b), the subset comprising one or more target nodes selected in dependence on the comparison.
A method according to any of the preceding claims wherein one or more of said probe measurements result from a plurality of probe test-messages being sent via a network path from a testing node (30) to the same target node at different times.
A method according to any of the preceding claims wherein the step of obtaining a plurality of probe measurements in respect of said first target node (30a) comprises obtaining a plurality of probe measurements resulting from probe test-messages being sent to the same target node, and wherein the step of comparing involves comparing said plurality of probe measurements resulting from probe test-messages being sent to the same target node with at least one of said one or more probe measurements resulting from one or more probe test-messages being sent to the other target node (30b).
A method according to Claim 10 wherein the plurality of probe measurements resulting from probe test-messages being sent to the same target node (30a) result from probe test-messages being sent to the same target node at different times.
A method according to any of the preceding claims wherein the probe measurements obtained in respect of the respective target nodes (30a, 30b) include measurements in respect of one or more of the following:
one or more characteristics associated with response time;

one or more characteristics associated with communication speed;

one or more characteristics associated with communication delay and/or delay variation;

one or more characteristics associated with communication volume;

one or more characteristics associated with reliability;

one or more characteristics associated with data loss;

one or more characteristics associated with communications quality;

one or more characteristics associated with security;

one or more characteristics associated with service usage.
Apparatus configured to perform a method in accordance with any of the preceding claims.
A computer program element comprising computer program code to, when loaded into a computer system and executed thereon, cause the computer to perform the steps of a method as claimed in any of Claims 1 to 12.